The study of radio galaxies selected at mJy flux levels has the potential to resolve two important issues in observational cosmology provided redshifts can be determined or reliably estimated for complete samples of such sources. First, the deep flux limit, combined with the shape of the radio luminosity function means that the redshift distribution of such samples provides a much more powerful test of the existence of a high-redshift cutoff for radio sources (Dunlop & Peacock 1990) than can be provided by further studies of brighter radio samples. Second, as a consequence of selection from bright radio surveys, the detailed study of galaxies at z > 2 has to date been confined to objects of extreme radio power (e.g. 4C41.17, Chambers et al. 1990; B2 0902+34, Eales et al. 1993), and it has now become clear that the ultraviolet-infrared properties of such sources are strongly contaminated by processes connected to the AGN (Eales & Rawlings 1993; Dunlop & Peacock 1993). Being 100-1000 times less radio luminous than these extreme sources, mJy radio galaxies at comparable redshifts should provide much more representative probes of the formation and evolution of elliptical galaxies in general.